Evaluation of Face Detection Algorithms for the Bank Client Identity Verification

Abstract

Results of investigation of face detection algorithms efficiency in the banking client visual verification system are presented. The video recordings were made in real conditions met in three bank operating outlets employing a miniature industrial USB camera. The aim of the experiments was to check the practical usability of the face detection method in the biometric bank client verification system. The main assumption was to provide a simplified as much as possible user interaction with the application. Applied algorithms for face detection are described and achieved results of face detection in the real bank environment conditions are presented. Practical limitations of the application based on encountered problems are discussed.

Analysis of results of large‐scale multimodal biometric identity verification experiment

“…Face image was acquired by RGB camera with resolution of 1920 × 1080 pixels at 30 frames/s rate. Extraction of facial image parameters was preceded by face detection in the scene [22], to focus only on the face location. The parameterisation procedure begins with 77 landmarks calculation on the detected face.…”

Evaluation of Decision Fusion Methods for Multimodal Biometrics in the Banking Application

“…All methods operate in feature spaces obtained by transforming the raw features or biometric signals to a representation more suitable for classification. In our work following approaches are implemented [ 28 ]: Face modality—the minimal distance in face keypoints descriptors 768 features space, composed of Histogram of Oriented Gradients and Local Binary Pattern features of 77 characteristic face landmarks, transformed with the Linear Discriminant Analysis [ 27 , 29 ], Voice modality—the maximum similarity of the speaker identity in the mel-cepstral frequency coefficients decision space statistically modeled with Gaussian Mixtures and Universal Background Models [ 3 , 30 ], Signature—accelerometer and a gyroscope signals processed with the triplet loss method, involving training a neural network to learn a new latent space representation, most suitable for maximization of the distance between signatures from different persons and minimization of the distance between signatures of the same person [ 31 ], 3D face image—the minimal distance between parameterized 3d meshes [ 32 ], Gaze tracking—the minimal distance in descriptor space, including statistical features of registered rapid eye movement speed (saccades), average, maximal, standard deviation, acceleration, length, distance, etc. Hand vein pattern—binary decision in a commercially available proprietary hardware unit by Fujitsu Identity Management and PalmSecure [ 3 , 33 ].…”

“…Face modality—the minimal distance in face keypoints descriptors 768 features space, composed of Histogram of Oriented Gradients and Local Binary Pattern features of 77 characteristic face landmarks, transformed with the Linear Discriminant Analysis [ 27 , 29 ],…”

Validating data acquired with experimental multimodal biometric system installed in bank branches

“…Extraction of facial image parameters is preceded by face detection in the scene. The face detection methods involved in the processing were presented in more detail in the previous paper by Szczodrak and Czyżewski (2017). The parameterization procedure is described more precisely in the paper by Bratoszewski et al (2017), nevertheless the main stages of the feature extraction process are recalled here.…”